Shaping the Future of Austria with Robotics and Artificial Intelligence

Vienna, November 2018

White Paper by the Austrian Council on Robotics and Artificial Intelligence Shaping the Future of Austria with Robotics and Artificial Intelligence

White Paper by the Austrian Council on Robotics and Artificial Intelligence

Vienna, November 2018 2 Austrian Council on Robotics and Artifi cial Intelligence

Preamble

The concept of the Austrian Council on Robotics and Artifi cial Intelligence was fi rst presented at the European Forum Alpbach 2017 by the Austrian Federal Ministry of Transport, Innovation and Technology and was constituted on 24 October 2017. At the time the White Paper was adopted, the Council had nine members:

Sabine Theresia Köszegi, Chairwoman of the Council

• Professor of Labour Science and Organization at the Institute of Management Sciences of TU Wien • Head of the MBA Program for Entrepreneurship and Innovation at TU Wien • Head of the Doctoral College "Trust in Robots" • Member of the High-Level Expert Group on Artifi cial Intelligence of the European Commission

Matthias Scheutz, Deputy Chairman of the Council

• Professor of Cognitive and Computer Science at the School of Engineering, Tufts University, Massachusetts, USA • Director of the Human-Robot Interaction Laboratory at Tufts University, Tufts School of Engineering (Massachusetts, USA) • Member of the AIWS Standards and Practice Committee of the Michael Dukakis Institutes for Leadership and Innovation of the Boston Global Forum • Member of the Partnership for AI

Mark Coeckelbergh

• Professor of Media and Technology Philosophy at the Institute of Philosophy at the University of Vienna • Professor of Technology and Social Responsibility at De Montfort University, Leicester, UK • President of the International Society for Philosophy and Technology • Member of the High-Level Expert Group on Artifi cial Intelligence of the European Commission

Corinna Engelhardt-Nowitzki

• Head of Department Industrial Engineering at the University of Applied Sciences Technikum Wien Shaping the Future of Austria with Robotics and Artifi cial Intelligence 3

Franz Höller

• Chief Technology Offi cer (CTO) KEBA AG, Linz • Chief Technology Offi cer KEBA AG, Linz

Sylvia Kuba

• Head of "Process Digitalization" at the Chamber of Labour Vienna

Andreas Kugi

• Professor for complex dynamic systems at the Vienna University of Technology • Board of the Institute of Automation and Control Technology (ACIN) and • Head of the Christian Doppler Laboratory for Model-Based Process Control in the Steel Industry • Co-Head of the Center for Vision, Automation & Control am Austrian Institute of Technology (AIT) • Vice President of the Austrian Association for Electrical Engineering (OVE) • Full member of the Austrian Academy of Sciences • Member of the German Academy of Science and Engineering (acatech) • Member of the Expert Panel of the Joint Science Conference (GWK) for the Excellence Strategy for the Promotion of Top-Level University Research in Germany

Martina Mara

• Professor for Robot Psychology at the Linz Institute of Technology (LIT) at Johannes Kepler University Linz

Erich Schweighofer

• Professor at the University of Vienna—teaches and conducts research in the fi elds of Legal Informatics, International and European Law • Head of the Centre for Computers and Law 4 Austrian Council on Robotics and Artifi cial Intelligence

This White Paper is the Council's fi rst written opinion. It is based on the technical status quo of robotics and AI at the constitutional period and defi nes the main fi elds of action and framework conditions for Austria from the Council’s point of view.

In accordance with its mandate, the Council is an advisory body set up by the BMVIT whose recommendations are publicly available to all stakeholders in politics, society, business and science. The Council seeks cooperation with other relevant bodies such as the Bioethics Commission, the Council for Autonomous Driving and the Advisory Council of the Digitisa- tion Agency.

Mandate of the Austrian Council on Robotics and Artiﬁ cial Intelligence

1 The Austrian Council on Robotics and Artifi cial Intelligence identifi es and discusses current and future chances, risks and challenges arising from the use of robots and autonomous systems (RAS) and artifi cial intelligence (AI) from a technological, economic, social and legal perspective.

2 The Austrian Council on Robotics and Artifi cial Intelligence thereby creates a strategic framework for BMVIT's own RAS and AI activities and drafts strategic guidelines. These guiding principles include, inter alia, opinions and recommendations on

I. creating economic framework conditions which promote innovation and technology and which ensure that the potential of RAS and AI are fully leveraged to ensure the competitiveness of Austrian industry;

II. creating a legal framework to ensure safe use of RAS and AI for individuals and society as a whole, in compliance with the legal framework of the European Union;

III. developing measures to identify, mitigate or prevent potential danger or harm to people and society caused by RAS and AI at an early stage; and

IV. planning public activities to inform the public on RAS and AI and to responsibly address society's fears and concerns.

3 The Austrian Council on Robotics and Artifi cial Intelligence regularly substantiates the strategic guidelines in the form of recommendations and implementation proposals to the Federal Minister for Transport, Innovation and Technology. Shaping the Future of Austria with Robotics and Artificial Intelligence 5 6 Austrian Council on Robotics and Artifi cial Intelligence

Executive Summary

The use of robotics and artifi cial intelligence (AI) in various areas of our lives will lead to fundamental changes in society. These new technologies have the potential to solve the great challenges of our times in many areas of application: dangerous, monotonous, unhealthy or strenuous activities can be performed by robotic systems; AI can contribute to better early diagnosis and treatment of diseases, and robots can support the autonomy and quality of life of the elderly and/or people dependent on care. Robotics and AI could help to sustain the international competitiveness of Austrian companies and thus create and secure jobs in the long term.

While discussing these chances we must, however, not forget about the ethical and social challenges that remain to be solved. These new technologies will not only entail changes and new requirements in the working environment: they will also raise ethical, legal and social questions that must be discussed and resolved by Austrian decision-makers from every area of society.

We have identiﬁ ed the following cornerstones for an Austrian robotics and AI strategy

1. Smart Governance All Austrians should beneﬁ t from robotics and AI technologies. We are convinced that a broad participation of all stakeholders—in particular of citizens themselves—in the process of defi ning this strategy is needed to increase the acceptance of new technologies. This can only be achieved if people's needs and fears are taken into account.

2. Smart Innovation We need a targeted research, development and investment policy to leverage the potential of robotics and AI technologies in all areas of application and thus tap into new markets and discover new fi elds of application. In doing so, the specifi c structure of Austrian economy, characterised by small and medium-sized companies, has to be taken into consideration. This will strengthen the economy and secure jobs and prosperity in Austria.

3. Smart Regulation A stable and secure framework is needed to ensure the trust of economic players and the positive development of markets. The use of robotics and AI must guarantee the safety of people and comply with ethical standards, fundamental human rights and European values. Wherever existing norms and standards are not suffi cient to ensure this, new European standards and norms must be developed without stifl ing socially benefi tial innovation. We need new and creative ways to resolve the trade-offs between innovation and regulation. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 7

In this white paper, we identify—based on the current state of research—four fi elds of action which we consider the top priorities for developing a smart strategy for robotics and AI:

Technology, R&D Workplace and Law and Society Awareness Raising, and Economy Qualiﬁ cation Communication and Public Relations

In these four fi elds of action, the White Paper addresses challenges and presents recommendations, which are to be understood as impulses for the initiation of a strategy creation process as summarized below:

Smart Governance

• Due to the complexity and pace of technological developments, we recommend an incremental strategy process ("strategizing") with institutionalized learning processes and feedback loops.

• The strategy process should include continuous research and technology monitoring.

• The Council recommends the broad involvement of stakeholders from all affected areas (e.g. research, technology development and production, business, education and training, health, safety, housing, mobility) in this strategy process.

• The Council recommends integration with the strategic activities currently taking place on the European level (European AI Alliance and AI High Level Expert Group).

• The Council recommends the wide-scale inclusion of Austrian citizens into the development of the Austrian Robotics and AI Strategy in order to inform and educate the public and to increase public acceptance. The Montreal Declaration on Responsible AI, for example, can be used as a reference. 8 Austrian Council on Robotics and Artificial Intelligence

Smart Innovation

• The Council recommends significant targeted public investment and measures to promote innovation in robotics and AI technologies and to develop infrastructure for the use of these technologies.

• At the same time, the Council recommends to introduce systematic training measures to ensure the availability of skilled professionals. In addition, extensive training programmes must be provided for the retraining of existing workforce.

• As major technical challenges are involved and the impact of these new technologies cannot yet be fully predicted, potential industries and application areas ("use cases") which permit rapid implementation and leveraging of robotics and AI potentials and enable a fast learning process are to be identified as quickly as possible.

• When selecting use cases, the following two criteria should be taken into consideration: framework conditions must be controllable, and a clear differentiation is needed between unproblematic and sensitive application areas. “Sandboxes" and “testbeds" are to be set up for sensitive and/or high-risk application areas, providing rapid learning cycles and knowledge transfer to stakeholders from all areas (research and development, business and politics).

• The experience gained from such use cases has to be continuously reviewed, so that after multiple feedback loops it can be used to lay down effective guidelines for government support measures and for development and innovation policy.

Smart Regulation

• The Council recommends reviewing the existing ethical and legal framework with regard to expected changes and, where appropriate, introducing new regulation and standards to ensure the safe use of robotics and AI for individuals and society. This must be done in close coordination with the current regulatory efforts of the European Commission.

• The Council recommends the establishment of appropriate certification, audit and compliance tools for robotics and AI technologies. Shaping the Future of Austria with Robotics and Artificial Intelligence 9 10 Austrian Council on Robotics and Artifi cial Intelligence

Table of contents

1. Introduction 11

2. Ethics as a Guiding Principle for the Council on Robotics and Artifi cial 15 Intelligence

3. Robotics and Artifi cial Intelligence – Defi nition and latest Research 17

3.1 Defi nition 17

3.2 State of Research 19

4. Social and Ethical Potentials and Challenges 23

4.1. Compliance with the European Union's Ethical Values and Principles 23

4.2. Safety and Security of Technologies 24

4.3. Human Oversight 25

4.4. Consideration of Social and Ecological Consequences 26

4.5. Initiating a Wide-Ranging Public Debate on Fundamental Questions 26 in Connection with Robotics and AI

5. Fields of Action 28

5.1. Technology, R&D and Economy 28

5.2. Workplace and Qualifi cation 31

5.3. Law and Society 35

5.4. Awareness Raising, Communication and Public Relations 40

6. Strategy Process and Governance 43

End Notes 44

Glossary 45

List of References 46

Online Resources 47 Shaping the Future of Austria with Robotics and Artifi cial Intelligence 11

1. Introduction

Robotics and Artifi cial Intelligence (AI) are strategically highly relevant technologies and involve the risk of disruptive change. It is therefore crucial that we take action now in Austria and shape the conditions that determine the world in which we want to live in the future.

With the existing knowledge about the course of technological change processes, we can prepare ourselves well for these changes. We will be able to shape innovation and technology in a way that not only makes our lives more comfortable, but also helps us meet the great challenges of today's world.

The developments in the fi elds of robotics and It is crucial to take initiative now AI, for example, have the potential to improve and to create a reliable the quality of life of many people with preven- framework that will shape our tion and precise diagnostics for all population future. groups; to increase the international competitiveness of Austria and to improve the quality of life of the population.

The aim is to strengthen the position of companies and to enable sustainable production and agriculture on the basis of more effi cient, environmentally and climate-friendly use of resources.

At the same time, technological progress also entails major challenges and risks. As a society, we can and must not leave the development of these important technologies to chance. Rat- her, we have to play an active part in shaping them.

In order to utilize the full potential of robotics and AI and at the same time ensure that all people can benefi t from these technologies, we need the joint efforts of politics, economy and society. In this context, innovation is not to be understood exclusively as technological progress, but also as the interaction of social, institutional and organisational innovation.

Austrian policymakers are called upon to invest in research and technological development, to create appropriate conditions and set incentives that on the one hand promote innovation and technical progress and on the other hand ensure that these developments do not lead to or exacerbate social inequality or jeopardise human rights.

Effective measures are to be taken in the coming years in the areas of research and technology promotion programmes, the defi nition of an appropriate legal framework and concepts for information, qualifi cation and integration of the population.

These must, of course, also be coordinated with ongoing strategic activities of the European Union. 12 Austrian Council on Robotics and Artifi cial Intelligence

The Austrian Council on Robotics and Artifi cial Intelligence was founded as an advisory body to support the development and implementation of such a robotics and AI strategy. It analyses the potentials, challenges, risks and effects of the use of robots and artifi cially intelligent systems. Its aim is to draft statements and recommendations for Austrian policymakers—proactively, if needed—which are practical and operational in real-life conditions.

The complexity of the issues and challenges at hand requires an interdisciplinary approach. Experts from various disciplines were therefore appointed to the committee - from robotics and automation, artifi cial intelligence, machine learning and computer science to law, social sciences, economics, psychology and philosophy of technology. Due to the expected far- reaching changes for the economy and the working world, representatives of the Chamber of Labour and the Federation of Austrian Industries are also represented in the Council on Robotics and Artifi cial Intelligence.

In order to guarantee responsible technology design and to sustain Austria’s competitive position, the Council is committed to the diversity of relevant perspectives and a holistic approach and interdisciplinarity: Council members’ diverse professional background enables them to analyse complex problems with an interdisciplinary, multi-perspective approach and formulating practical recommendations, simultaneously using specifi c competences.

These recommendations are not limited to the Federal Ministry of Transport, Innovation and Technology’s more narrowly defi ned agenda on technology policy, but are addressed to all ministries concerned and all fi elds of action of the Austrian Federal Government.

In addition, the Council endeavours to create synergies with existing Austrian, European and international bodies and interest groups from the fi eld of new technologies and digitisation; include their expertise and to conduct a productive dialogue.

On 25 April 2018, the European Commission published a Communication on "Artifi cial Intel- ligence for Europe" (European Commission, 2018), announcing the development of a Euro- pean Strategy on Artifi cial Intelligence, which would put the EU at the forefront of robotics and AI. The document outlines a set of actions for the coming years. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 13

The objectives of the European AI Initiative are

strengthening the EU's technological and industrial performance

encouraging the adoption of AI in all areas of economy

preparing for the socio-economic changes associated with AI

establishing an appropriate ethical and legal framework

In the view of the Council, these objectives of the European AI Initiative can also be adopted as objectives for an Austrian robotics and AI strategy. The Austrian Council on Robotics and Artifi cial Intelligence considers it its task to actively shape the exchange of information and cooperation with EU expert committees for Artifi cial Intelligence and to support the Austrian government in proactively participating in European robotics and AI policy.

The Council regards its work as an essential The Council regards its work as an contribution to shaping a sustainable future in essential contribution to shaping a which technology and humans are not played sustainable future in which off against each other, but are understood as technology and humans are not complementary factors. Robotics and AI are played off against each other, but intended to improve people’s quality of life, are understood as complementary enhance their skills and increase their auto- factors. nomy. To achieve these goals, the Council on Robotics and Artifi cial Intelligence relies on fundamental ethical principles.

They are outlined in the following section. Section 3 contains a more detailed defi nition of robotics and AI, illustrating their relevance and problems. In addition, it includes a brief overview of the state of research and prognoses with regard to the further development of research in the fi eld of robotics and AI. In section 4, fundamental potentials and challenges in connection with robotics and AI are presented. Section 5 identifi es the key areas where the Council on Robotics and AI sees the need to act and which therefore should be at the focus of an Austrian Strategy for Robotics an AI. By setting such priorities, the Council intends to help defi ne key initiatives for Austria. 14 Austrian Council on Robotics and Artificial Intelligence Shaping the Future of Austria with Robotics and Artifi cial Intelligence 15

2. Ethics as a Guiding Principle for the Council on Robotics and Artiﬁ cial Intelligence

The Austrian Council on Robotics and Artifi cial Intelligence follows accepted ethical principles, and recommends their application to national and international decision-makers.

This includes in particular the following ethical principles, values and rights:

Human rights: enshrined in national and international charters of fundamental rights, human rights treaties and declarations, they cover human freedom and dignity, economic, cultural and social rights and the protection of privacy;

Justice and fairness, inclusion and solidarity, including the protection of vulnerable members of society;

Democracy, social and political participation;

Principle of non-discrimination;

Personal, social and shared2 responsibility;

Additional ethical and social values in Austria and Europe3

Due to its mission, the Council also needs to consider additional standards when drawing up recommendations and statements: In addition to the general safety and development standards recommended by standardisation authorities and international expert groups (e.g. "Institute of Electrical and Elecronics Engineers"—IEEE), the principles of "Responsible Innovation" and "Ethically Aligned Design” of the international IEEE initiative must also be observed.

These initiatives call for a proactive approach to technology development that takes ethical and social considerations into account from the early stages of innovation and design and involves relevant stakeholders in the development process at an early stage.5

This process has to be participatory6 to ensure that society as a whole can benefi t from these technological developments. Only a participatory approach can ensure that the benefi t of society as a whole is not outweighed by particular interests.

In addition, the Council relies on the general principles of research ethics7, with a special focus on the role of interdisciplinary and transdisciplinary cooperation in research and development. Cooperation between various disciplines and stakeholders from business, society and research are to be specifi cally promoted. 16 Austrian Council on Robotics and Artificial Intelligence Shaping the Future of Austria with Robotics and Artifi cial Intelligence 17

3. Robotics and Artiﬁ cial Intelligence – Deﬁ nition and latest Research

3.1. Deﬁ nition

Robots8 are machines that perform tasks–to some extent or fully–autonomously. In the light of the increased cognitive abilities that these machines have acquired with the use of intelligent software, they are now often referred to as "autonomous cognitive systems". The cognitive aspects of such systems include automatic data processing, creating environmen- tal models and target-oriented action plans, the decisions made and action taken based on these models, and the automatic improvement of system functions using machine learning.

Autonomous cognitive systems are for example collaborative robots in industry and services, automated vehicles and other autonomous transport systems, active exoskeletons, social robots or medical robots as well as cognitive assistance systems and autonomous software agents.

An essential component of autonomous cognitive systems is “Artifi cial Intelligence". For this reason, the terms "AI systems" or "cognitive systems" are often used synonymously. The European Commission’s defi nition of AI (2018) illustrates that a clear distinction is hardly possible:

Artificial intelligence (AI) refers to systems with an "intelligent" behaviour which analyse their environment and act with a certain degree of autonomy to achieve certain goals.

AI-based systems can be exclusively software-based, operating in a virtual environment (e.g. language assistants, image analysis software, search engines, speech and facial recognition systems), but can also be embedded in hardware systems (e.g. modern robots, autonomous cars, drones or "Internet of Things" applications).

We use AI every day to translate texts, create subtitles in videos or block unwanted emails. Many AI applications need data to function optimally. Once they are functioning smoothly, they can improve and automate decisions.

Being aware of the rapid pace of technological change, the Council generally refers to "autonomous cognitive systems" to include both robotics and AI systems. This terminology is open to further clarification on a case-by-case basis. 18 Austrian Council on Robotics and Artificial Intelligence

AI research can be roughly divided into knowledge-based symbolic and data-based sub- symbolic approaches, although hybrid approaches have been gaining ground recently. Knowledge-based systems such as Classical Cognitive Systems have explicit symbolic representations of available knowledge, which they utilise to make decisions.

This knowledge is partly provided as default, but the system is also able to gain new knowledge from the tasks it is performing (e.g. through verbal instructions, by reading texts or by observing events in their environment).

Currently, data-based systems are mainly developed at the sub-symbolic level, such as deep neural networks. They acquire implicit knowledge in a long “training phase" using machine learning methods, before they are ready to perform their designated tasks. This means that such systems depend on the availability of high-quality data that can be used to train them. Hybrid systems use the advantages of explicit knowledge representation (e.g. for natural language interaction of robots with humans) and data-based methods (e.g. for image recognition of autonomous vehicles).

Artificial Intelligence Research

knowledge-based data-based (statistical) approaches symbolic approach Machine Learning

Neural Networks

Deep Learning

Fig. 1: Approaches in AI Research (see Perset, Nishigata & Carblanc, 2018)

It is generally expected that the recent successes of AI will continue and intensify in the future. This raises the fundamental question whether it is possible to replicate or even surpass human intelligence and cognition in machines—with all the ethical consequences that such a development would entail. However, a long-term prognosis is difficult, as (disruptive) innovation in this field is happening at unprecedented speed. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 19

Analyses of errors, ﬁ ndings and lessons learned from previous prognoses on AI made between 1950 and 2012 show that

There is no convergence and only little correlation between expert prognoses,

there is no indication that prognoses made by experts are more accurate than those of laypersons, and that

there has been a strong trend towards long-term forecasts (15 to 25 years from the forecast date) on AI development, and that these predictions have been far to optimistic.

These fi ndings should warn us of making ill-considered predictions about the future that could unnecessarily fuel fears and lead to unwarranted scepticism towards new technologies.

Although signifi cant progress has been made in AI development, adaptive autonomous sys- The form of intelligence associated tems are still limited to very specifi c abilities, with human beings cannot be even if they can sometimes surpass humans achieved yet with technologies and in this narrow framework. In general, people AI algorithms currently available. are referred to as "intelligent” because they can comprehend new situations and solve problems. This form of “general intelligence” is not yet achievable with currently known technologies and algorithms for AI systems.

3.2. State of Research

In recent years, research in the ﬁ eld of robotics and AI has been fostered by

the widespread availability of affordable powerful computer hardware (computing power, data storage),

intelligent and networked sensors, sensor data fusion,

new actuator technologies and control algorithms,

global networks with high bandwidths and a multitude of data sources (social networks, online platforms, Internet of Things), as well as

the development of new powerful algorithms

and has already led to signiﬁ cant scientiﬁ c breakthroughs and demonstration systems. 20 Austrian Council on Robotics and Artifi cial Intelligence

The active participation of major technology companies such as IBM, Microsoft, Google, Facebook has contributed signifi cantly to the spread and refi ning of AI in recent years, as these companies have been able to provide both the fi nancial resources and the large amounts of data required for training machine learning algorithms.

In 2011, for example, IBM’s Watson was able to defeat the best human Jeopardy player for the fi rst time, thereby proving that intelligent systems can conduct effective question-and -answer dialogues in natural language. Shortly thereafter, Google introduced Alpha-Go, a program that defeated the best Go player in the world in 2016, as well as the follow-up program Alpha-Zero in December 2017, which despite minimal requirements could learn to play chess within a few hours at the level of a world champion.

AI research is subdivided into various fi elds, forming largely independent scientifi c com- munities in areas like machine image or language processing, machine learning, logical reasoning, planning and testing, dialogue systems (e.g. the new Google Duplex System, which can solve simple everyday tasks independently in natural language dialogues), game theory, decision theory, knowledge representation, cognitive systems and integrated control architectures.

The methods used can be roughly divided into symbolic and sub-symbolic approaches:

Symbolic methods include both logic-based and stochastic approaches; while the subsym- bolic methods are usually represented in the form of neural networks and more recently by so-called deep neural networks. The latter have been able to achieve astonishing successes, in particular in areas like image processing and in learning simple control rules (by "deep Q-learning"). Some of these AI systems are already so advanced that they are clearly superior to human performance in certain tasks in terms of speed and result quality.

While logic-based methods are frequently The latest AI systems are superior used in language comprehension, formal ver- to humans regarding certain ifi cation, semantic networks and general prob- precisely deﬁ ned tasks. lem solving; stochastic methods are mostly applied in robotics, where signals are typically noisy. Therefore, absolute values are not an option—distributions have to be used instead.

In addition to AI, vast progress has also been made in robotics development, for example in humanoid robots or so-called lightweight robots. For example, the research company Boston Dynamics managed to develop a humanoid robot which can autonomously pick up heavy objects from the fl oor and sort them into a shelf. What is more, after doing a backfl ip, the robot can stand on both legs without falling over. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 21

A rapid development is also taking place in classical industrial robotics, complementing the previous generation of rigid, highly specialized machines by fl exible and adaptive robot and tool systems with specialized cognitive capabilities. This development is driven, among other things, by the increasing individualisation of mass-marketed products and the associated increasing demand for fl exibility in production.

There is already a large number of so-called lightweight robots available on the market that are limited in their payload, but have been designed to interact safely in direct contact with humans and can be operated easily and intuitively.

Also, research in recent years has made autonomous drones far more agile, with vastly improved manoeuvrability and countless additional functions. For example, state-of-the-art technology allows the targeted use of stalls to navigate drones through narrow openings.

The integration of robotics and AI can best In classical industrial robotics, a rapid be illustrated by the development of autono- development is taking place: the previous mous vehicles, which combine modern con- generation of rigid, highly specialized trol technology with machine perception and machines is now complemented by ﬂ exible advanced planning, simulation and decision and adaptive robot and tool systems algorithms in order to be able to navigate with specialized cognitive capabilities. autonomously without collisions in a highly dynamic environment, such as innercity road traffi c.

Other potential application areas range from real-time language translation to diagnostics based on global databases, that enables early detection of diseases; and from optimisation and customisation (e.g. in buildings), maintenance and resources to individual learning. AI and robot technology are already increasingly used in digitally networked production and the social sphere e.g. as assistance robots in hospitals or surveillance robots in shopping centres.

In addition, robots are also used in households: from vacuum cleaner robots to social robots such as Jibo, which can—similarly to other natural language systems like Amazon’s Alexa—complete smaller tasks (such as display photo albums, search online, schedule appointments etc.).

New and more powerful variants of these "intelligent cognitive assistants" are to be expected in the near future. 22 Austrian Council on Robotics and Artifi cial Intelligence Shaping the Future of Austria with Robotics and Artifi cial Intelligence 23

4. Social and Ethical Potentials and Challenges

Robotics and AI not only have the potential to make our lives more comfortable, but also to help us solve the great challenges of modern society. A few examples to illustrate key developments: robotics and AI deployed in the health sector will help to signifi cantly improve early detection and treatment of illnesses. Nursing assistance systems could support high quality care and thereby bolster the autonomy and quality of life of people in need of care.

Robotics and AI technologies could take over dangerous, monotonous, repetitive, physically unhealthy or strenuous activities from humans (such as inspection work in mines or clean- up work in radioactively contaminated areas).

Robotics and AI could help us solve Deploying robotics and AI could secure the the great challenges of modern international competitiveness of companies society. and thus maintain jobs in a high-wage country like Austria in the long term. More effi - cient, environmentally and climate friendly use of resources—for example in agriculture and industry—are the key to securing sustainable prosperity.

While discussing these opportunities we must not, however, forget about the ethical and social challenges that remain to be solved. The Council recommends that political decision-makers involved in the development of a Robotics and AI Strategy for Austria pay particular attention to the following principles in addition to the technological and economic aspects:

4.1. Compliance with the European Union’s Ethical Values and Principles

In developing and using robots and AI systems, general ethical principles must be observed. Therefore, it must be ensured that the deployment of robots and AI systems safe- guard the principles of human rights, including data protection and privacy, as well as the related principles of human dignity, human freedom and autonomy.

A clear recommendation of the Council is to AI systems design has to ensure ensure that AI systems—as far as technically transparency: AI decisions should possible and as far as this contributes to be comprehensible for those the protection of users and affected per- affected. sons—are designed transparently and that decisions taken by AI systems made comprehensible (“explainable AI“). Experts are currently discussing the implementation of an “Ethical Black Box” 10 and alternatively the “Counterfactual Explanations“ approach11. 24 Austrian Council on Robotics and Artifi cial Intelligence

Compliance Safety and with ethical values security of and principles technologies

Robots & Consideration AI systems Responsibility of social remains with and ecological humans consequences

Strong civil involvement and public participation

Fig. 2: Fundamental Principles for Developing a Robotics and AI Strategy

In addition, more far-reaching values and general principles such as justice and fairness, diversity and inclusion, solidarity and protecting vulnerable people must also be taken into account in the course of developing and deploying new technologies.

4.2. Safety and Security of Technologies

It must be ensured that robotics and AI technologies are safe to use, and comply with general development standards. Key questions in this context are whether there is a need to adapt existing standards and whether new standards are needed.

The safety and security of new technologies includes both machine safety and IT security, as well as the integration of these aspects. Attention should also be paid to enabling a high subjective sense of safety in using robots and AI systems. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 25

4.3. Human Oversight

Although robots and AI systems are becoming more and more intelligent, their moral status is still unclear. In practical applications, the question arises whether robots and autonomous AI systems can be moral agents (in a practical, not philosophical sense) both able and authorised to make ethical decisions on their own. This aspect is discussed, for example, in connection with autonomous vehicles, as they might be involved in situations where they have to make ethical decisions. Such decisions could range from breaking laws to avoid an accident (e.g. ignoring a stop sign in order to avoid a rear end collision) to deciding which life is more worth protecting in the event of an unavoidable collision (e.g. that of the passengers or of other road users).

In view of the increasing autonomy of such Human oversight and a clear moral systems and the associated delegation of and legal framework must be decisions to technical systems, it is neces- deﬁ ned for the increasing delegation sary to ensure human oversight in every sit- of decisions to AI systems. uation and to defi ne a clear moral and legal framework which also clearly outlines the responsibilities of all parties involved (e.g. developers, manufacturers, operators, users, customers, etc.).

Important cornerstones were set out by the European Parliament’s12 resolution of 16 February 2017 with recommendations to the Commission on Civil Law Rules in the Robo- tics Area (2015/2103(INL) (report compiled by: Mady Delvaux) and with the preparatory study13, European Civil Law Rules on Robotics by Nathalie Nevejans, whereby an EU-wide regulation of liability issues and an ethical code of conduct were called for.

It is important to note here that this Council Machines cannot and should not recommendation should not be regarded assume moral responsibility. as a carte blanche for the development of Ethical responsibility for robotics autonomous machines without any ethical and AI systems must ultimately regulatory mechanisms. On the contrary: the remain with humans. Council recommends that algorithms and computational precautions for ethical behaviour be directly integrated into machine architecture and machine design in order to minimize possible ethical violations and amoral behaviour in autonomous machines. 26 Austrian Council on Robotics and Artifi cial Intelligence

4.4. Consideration of Social and Ecological Consequences

Robotics and AI technologies must be aligned with human needs; respect human behaviour and human diversity and must not enforce any inhuman or inhumane adaptation of humans to technology. What is more, it must be ruled out that technologies lead to biased or discriminatory treatment of individuals.

It is crucial to assess potential economic and Robotics and AI technologies must employment impact of new technologies at be aligned with human needs. an early stage, and to develop technologies that support sustainable economic activity that is both economically and ecologically benefi cial.

4.5. Initiating a Wide-Ranging Public Debate on Fundamental Questions in Connection with Robotics and AI

It must be ensured that political decision-makers already today refl ect fundamental social questions in connection with robotics and AI which might not be relevant immediately, but which we will be confronted with in the medium term. Experts from various disciplines, interest groups and the general public are to be involved on a meta-level in a comprehensive and interactive discourse on fundamental issues. Political decision-makers must now begin to reﬂ ect the fundamental and This includes, for example, a fundamental long-term effects of robotics and AI. debate on how AI technologies can infl u- ence our democratic system and how we can uphold democracy and the rule of law in the long term. We also need to discuss how work—paid and unpaid—is to be distributed between humans and machines in future. This also requires a new understanding of the concepts of work and labour. The latter is closely connected to challenges associated with people in need of care. Robotics and AI will undoubtedly play an important role in addressing these issues. However, we have to clarify fi rst which roles we want to assign to these technologies. Shaping the Future of Austria with Robotics and Artificial Intelligence 27 28 Austrian Council on Robotics and Artifi cial Intelligence

5. Fields of Action

Based on latest research, the Council on Robotics and Artifi cial Intelligence has prioritised several fi elds of action in a detailed discussion. These priorities are meant to serve as a basis for the development of an Austrian Strategy for Robotics an AI. We see an urgent need for action in the following four areas:

Technology, Workplace and Law and Awareness Raising, R&D and Economy Qualiﬁ cation Society Communication and Public Relations

These four fi elds of action are closely inter-connected, therefore it is not always possible to draw a clear line between them.

5.1. Technology, R&D and Economy

An internationally competitive and "fi t for the future" economy is the basis of Austria's past and future prosperity. This position can only be maintained if a framework is created that allow companies to succeed in global competition.

Tackling digitisation requires Austrian economic stakeholders to be willing to innovate and invest.

However, they will only do so if they can expect a good return on their investment. In a high-wage country like Austria, companies must occupy creative niches and build on their existing strengths to succeed on global markets. Areas in which both the academic fi eld (knowledge generation) and the commercial fi eld (knowledge utilisation) are already strong are to be consolidated and consistently developed with the help of robotics and AI. To this end, both the potential of small and medium-sized enterprises (SMEs) and large corporations must be equally accounted for by political decision makers.

The foundations of prosperous development must be provided, ranging from effi cient infrastructure to the creation of business-friendly and innovation-promoting conditions. This also includes access to data as a business resource. Political stakeholders have to ensure a high degree of legal certainty for business activities, while protecting the human rights for individuals. All in all, these basic principles are key factors of competitiveness, which the Council believes should by all means be incorporated into all future measures. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 29

Austria has various research institutions that focus on robotics and AI, ranging from basic research to the development of marketable prototypes. They cover a wide range of topics, from motion control of manipulators and mobile systems to the perception and understanding of environments; from human-machine interaction to machine learning and AI. Latest studies provide more detailed information14.

Although there are individuals offering outstanding expertise in their respective fi eld, Aus- tria lacks an overarching strategy with a clear vision15.

In view of the limited size of the Austrian research landscape and the scarcity of available human and fi nancial resources, it is essential to defi ne clear priorities and focus on application areas which are relevant for Austria.

In view of the limited size of the Austrian re- A focused state funding and search landscape and the scarcity of available investment policy could provide the human and fi nancial resources, it is essential necessary infrastructure and set to defi ne clear priorities and focus on applica- clear priorities in ﬁ elds of application tion areas which are relevant for Austria. relevant to Austria. It is essential to involve key businesses into the strategy process right from the start. Continuity of innovation chains from basic research to successful industrial implementation must also play an important role in the strategy process.

In the future, it is to be expected that robotics and AI technologies will be deployed in almost every area of life.

Digital assistance systems, which facilitate human decision-making by providing information that is generated from data according to a given set of rules, are the fi rst step. In other areas, however, AI systems should be enabled to operate autonomously, i.e. to solve tasks independently, to make independent decisions within a given framework, and to react ade- quately to unforeseeable events.

The timeframe within which such developments are expected to be implemented depends on various factors and varies from application to application. In general, however, it is safe to assume that tasks requiring a high level of physical dexterity, fl exibility and sensitivity in addition to cognitive abilities will be more diffi cult to solve than tasks that can be performed with purely software-based systems.

Signifi cant growth rates are predicted in all robot application areas, with "social robots" in particular expected to achieve very high growth rates. However, these developments also entail some hazards. In addition to classical safety problems with robots, such as collsion-free navigation or the danger of manipulation, social aspects must be taken into consideration too. Machine learning algorithms could potentially be dangerous, as it is 30 Austrian Council on Robotics and Artificial Intelligence

currently not clear which categories these algorithms can extract from data sets. For example, it is known that deep neural networks trained on image recognition of traffic signs for applications in autonomous vehicles are completely misled by even a slight modification of the image, for example by a sticker on a traffic sign, and can no longer recognize a stop sign as such.

The question of the extent to which the behavioural characteristics of autonomous robots must be formally specified and verified before they can be used in practice is being intensi- vely discussed by a wide variety of interest groups and requires further research, but also a clear legal framework. At the end of the day, Austria and the EU have to formulate their legal positions in order to guarantee the safety of autonomous systems for their citizens.

For a high-tech country like Austria, that has highly specialised professionals, disruptive changes such as digitisation offer the chance to be at the forefront of key technologies. As a very small research and development community by global standards, however, Austria should not measure itself against the ambitions of big nations (China, USA, Japan, Germa- ny...), but—as part of an international network—should occupy niches in research on these key technologies.

Therefore, the Council recommends targeted public investment of significant amounts, measures to foster innovation in robotics and AI technologies, and provision of the infrastructure needed for the use of these technologies. The question of using data collected by public authorities (e.g. health data made available for research purposes) must be discussed and an appropriate strategy has to be developed, which takes into account the interests of citizens’ worthy of protection.

The Council recommends to define key application areas ("use cases"), relevant fields of technology and future priorities in R&D as soon as possible. In doing so, it is to be ensured that investments permit rapid implementation and realisation of the potentials of robotics and AI, and enable a rapid learning process.

When selecting use cases, the controllability of the framework conditions and risk must be taken into consideration. A distinction between unproblematic (no risk) and sensitive areas (high risks) of application should be made. “Sandboxes" and “testbeds" should be set up for sensitive, high-risk application areas and/or if the impact of these new technologies cannot yet be safely predicted.

The aim of these measures is to create fast learning processes and knowledge transfer to stakeholders from all backgrounds (research and development, business and politics).

The experience gained from such use cases has to be continuously reviewed, so that after multiple feedback loops it can be used to create effective guidelines for government funding, development and innovation policy. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 31

5.2. Workplace and Qualiﬁ cation

Robots and AI systems are increasingly able to perform complex cognitive activities that previously only humans were capable of.

The ever-stronger intertwining of virtual and physical processes in all areas of life—humans, machines, organisations, businesses, society or the state—change life and work situations and require new ways of thinking, working practices, new ways to cooperate and additional skillsets.

First and foremost, this means basic digital literacy of the general public and extends to specifi c expertise in the development, use and optimisation of intelligent machines for a given purpose. In addition, innovation and creativity, logic and integrated thinking will become more relevant in contrast to pure factual knowledge. Empathy, social, intercultural and communication skills play an increasingly important role with regards to human-machine interfaces and the evaluation, selection and sharing of results provided by digital assistance systems.

In light of these factors, the Austrian Council on Robotics and Artifi cial Intelligence regards the working world of the future as a key fi eld of action. The key question is how opportunities for the labour market can be leveraged and how the working conditions of employees, work organisation and income development can be addressed.

It is still not fully clear what effects robotics and AI will have on the future working world. Forecasts on the potential impact of of new technologies on the labour market vary widely, ranging from predicting enormous job losses to positive effects on employment. Very pes- simistic expectations concerning labour markets can be found, for example, in Ford (2015), Brynjolfsson and MacAfee (2011) or Frey and Osborne (2017). These authors predict extensive automation of routine tasks in almost all industries and expect almost every second to be job affected. Although they are frequently quoted in the media, these forecasts are highly controversial among experts because the forecasting models used do not distinguish between individual tasks and entire professions or job profi les.

Therefore, it is safe to assume that automation potentials tend to be overestimated. Rigidly structured tasks such as routine workfl ows are indeed easy to automate. In contrast, less structured tasks cannot be easily taken over by machine systems..

In any case, it can be assumed that simple The latest IHS and OECD forecasts physical activities in particular, which can be see 9% to 14% of jobs—in particular easily completed by robotics and AI, will be those involving simple activities that taken over by technical systems, depending can be easily taken overby robotics on the economic and legal framework and and AI—at risk. cultural acceptance. 32 Austrian Council on Robotics and Artifi cial Intelligence

Recent OECD forecasts see about 14% of all jobs at risk. The Institute for Advanced Studies (IHS, 2017) predicts that 9% of jobs in Austria will be automated. Other studies predict the creation of new jobs, particularly in connection with the development, optimisation and maintenance of new systems.

According to the latest report of the World Economic Forum (2018), corporations expect a steep increase in new tasks and fi elds of work as a result of digitisation. At the same time, it is clear that every second member of the workforce will need signifi cant additional training and retraining within the next four years. The Council also sees an urgent need for education, training and retraining in Austria.

Projected average need for retraining in enterprises commensurate to the number of employees, 2018–2022

Retraining needs of less than 1 month, 13%

Retraining needs of 1–3 months, 12% no need for retraining, 46% Retraining needs

Retraining needs of 3–6 months, 10%

Retraining needs of 6–12 months, 9% Retraining needs of over 1 year, 10%

Fig. 3: Forecast on Retraining Requirements (see WEF, 2018)

A redistribution of paid work can already be observed today. However, a detailed examina- tion reveals clear differences depending on the level of income and education. While in the past mainly simple manual tasks requiring low skill- levels were automated, now cognitive robotics and AI systems can also automate tasks that previously required a higher level of education and were therefore performed by so-called knowledge workers.

In certain sectors, such as banking or retail, automation is already very advanced on the level of administrative staff. Similar developments can be expected in other industries. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 33

The economist David Autor (2015) assumes that paid work will not decrease overall, but a polarization effect might occur. Robotics and AI create jobs for well-educated people with a specifi c skillset: analytical abilities, abstraction, ability to plan and good judgment, orien- tation and problem-solving skills, creative intelligence and social skills such as empathy and selfrefl ection are required. On the other hand, there will also be a large number of tasks which include simple manual activities that cannot be automated, or for which automation is not economically viable.

Deploying low-skilled workers, who are enabled to perform more complex tasks by intelligent assistance systems, could also contribute to this polarisation effect.

This could lead to more precarious work (job insecurity) and increase social inequalities. Appropriate measures have to be taken to prevent increased job insecurity. On the other hand, digital assistance systems could help preserve the jobs of low-skilled workers, which are often considered to be at risk16.

There is a thin line between risks and oppor- Appropriate measures have to be tunities - as is often the case with new tech- taken to prevent increased job nologies that have unpredictable consequen- insecurity due to robotics and AI. ces. It is therefore essential to actively shape these developments to secure employment and high working standards in Austria.

Overall, it is a challenging and important task to implement measures for the qualifi cation of workers and social policies, which take into account the qualitative effects on working conditions, work organisation and forms of work as well as on job structures. In fact, the OECD (Nedelkoska & Quintini, 2018) sees a high probability of change in around 32% of all jobs in terms of work organisation.

It is important to identify risks - for example with regard to the physical or mental health of workers or safety at work. However, it is also necessary to understand how robotics and AI applications could improve working conditions. Therefore, best practice examples and research on people in the work process are of particular interest. Equally important is the integration of employees' experience and knowledge into the design of new working environments.

The Council regards close monitoring of these developments as a key task in order to ensure that society a whole can benefi t from these changes. Short, medium and long-term measures are to be developed in order to cope with the expected changes in the working world. Training and re-training measures are the fi rst priority in the short term. Industry representatives stress that the prevailing lack of skilled workers is inhibiting growth. 34 Austrian Council on Robotics and Artifi cial Intelligence

From a technical point of view, professional education and training should focus in particular on mechanical engineering, mechatronics, automation technology, electrical engineering, mathematics, information technology and statistics. In addition, the high complexity and strong integration of machines requires interdisciplinary skills in systems engineering and project management, process analysis and design skills, and the ability to integrate new technologies into a given application scenario.

In view of the rapid pace of technological development, skills required for future jobs will be constantly changing. Therefore, the willingness to learn, fl exibility, and the ability to assess unclear cause-effect scenarios and elasticities (heuristically or probabilistically) are required at all qualifi cation levels.

It is clear that basic skills for using digital In future, Austria's population will technologies, robotics and AI are required for need basic digital literacy and all member of society. In addition, specialist competences in using robotics and knowledge on the use and shaping of these AI systems. technologies is required. This starts with the basic MINT competences (competences in mathematics, computer science, natural sciences and technology) required and ran- ges from the methodological tools of technicalscientifi c domains to practical knowledge and skills in technology areas such as data science, robotics, machine learning, etc.

In contrast to basic digital skills, these advanced technological skills will be required to var- ying levels. In addition, there will be an increasing need to strengthen human-centric skills such as interdisciplinary discourse, cooperation, willingness to learn and fl exibility. With most areas of life and work being digitised, individuals and society as a whole have to learn to use new technologies.

Particular attention must be paid to gender- Initiatives that in particular awaken in- relevant aspects when developing a suitable terest among girls and women for tech- education and training strategy. Initiatives nical, mathematical and natural science that inspire girls and women in particular to subjects in the education system and for study technical, mathematical and scientifi c technical professions must be intensiﬁ ed subjects and to work in technical profes- and expanded. sions should be fostered and intensifi ed.

Existing barriers - both implicit and explicit - to career development in the education system and in professional life are to be eliminated. Encouraging and integrating women into technical professions is regarded by the Council as an essential factor not only in averting the looming lack of skilled workforce, but also for ensuring the sustainable and responsible development of robotics and AI.

In view of the rapid technical change, an appropriate qualifi cation strategy must also take into account the responsiveness of the education and training system, as well as opportu- Shaping the Future of Austria with Robotics and Artifi cial Intelligence 35

nities for support provided by active labour market policies. This also means that education and training should not be confi ned to formal institution. Other stakeholders in the education and training system—e.g. businesses and interest groups—must also be involved. Improved training for teachers with modern teaching and learning methods is also required, as they are the ones to guarantee the quality of future education and training.

Right now, it is also essential to conduct a detailed analysis of the medium- and long-term effects of technology use:

In addition to the fundamental questions mentioned above on shaping and distributing work in future, socio-political considerations must also be taken into consideration.

How can social policies help those, who are In addition to public institutions, other (for whatever reason) excluded from employ- stakeholders such as businesses and ment? This is another challenging question interest groups also need to take that needs to be addressed. Appropriate mea- part in providing education and sures have to be taken to enable exchange of training to all members of society. information and an open and inclusive debate free of ideological or political party’s interests on these fundamental questions—in Austria and with other European countries.

5.3. Law and Society

In addition to their use in production, manufacturing and other highly specialized work environments (international space stations, operating theatres, etc.), autonomous robots and AI systems are increasingly being used in social contexts such as nursing and health care, psychological counselling (online or over the phone), customer care and telemarketing.

Robotics and AI technologies are also ex- Robots and AI technologies will also pected to play an increasingly important role play an increasingly important role in the leisure sector. Digital, artifi cially intel- in the leisure sector. ligent assistance systems (Amazon’s Alexa, Google Assistant etc.) as well as cleaning or mowing robots are already showing considerable user numbers in the private customer segment. So-called “personal robots" or “companion robots", which have a mobile physical shape and can provide simple assistance and entertainment services (e.g. taking photos, playing games), are not yet widespread. However, they are expected to become more widely available and more affordable in the coming years.

Unlike industrial robots, which perform clearly defi ned tasks in highly specialized work areas, the functionality of these social robots is for the most part not restricted to a specifi c environment or task. 36 Austrian Council on Robotics and Artifi cial Intelligence

Robotics & AI Systems

Industrial Robotics Professional Service Robotics Robotics in Households e.g. assembly robots, e.g. transport robots, operation e.g. cleaning robots, assistance manipulators, sorting robots, robots, care robots, guard robots, robots, entertainment and toy cyber-physical systems, rescue robots, telepresence robots, robots as companions Industry 4.0, collaborative robots, information robots (Companion Robots) robots (cobots)

Fig. 4: Application Areas for Robotics

In addition, their appearance is often similar to that of humans or animals and they are equip- ped with abilities for natural language communication and for simulating human emotions or human character traits. Anthropomorphic design features such as faces, eyes and arms are used to transmit non-verbal signals known from interpersonal communication to the robot, thus promoting a "social" connection with the machine.

In fact, relevant social science literature suggests that people—depending on personal and situational factors—tend to perceive intentionality or other human characteristics in robots and AI systems, identify with them, sympathize with them, show affection and trust, or even lie for them. These effects occur particularly with humanoid robots, but also sometimes with non-humanoid robots (e.g. vacuum cleaner robots). At the same time, numerous empirical fi ndings indicate that highly humanoid robots often trigger strong adverse reactions17 and that society is widely sceptical about robots that assume social communicative functions perceived as "typically human”18.

The Council suggests to monitor research on the acceptance of different "social", "emotional" or "humanoid" robots and suggests to take state-of-the art fi ndings for policies into consideration. The Council also points out the need for further research into the psychoso- cial, social and societal implications of such “humanoid” robots.

Apart from the exemplary developments in social robotics, questions of human-machine interaction and technology acceptance are also increasingly relevant in other contexts. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 37

The current trend towards collaborative ro- Humans tend to perceive human botics, the aim of which is close coopera- characteristics in robots and AI tion between humans and machines in the systems, to identify with them and workplace, or the expected deployment of to show compassion, affection and autonomous robots in public spaces (e.g. trust. autonomous cars, transport robots, drones) leads, among others things, to the phenome- non of very different people, sometimes with little affi nity for technology, increasingly coming into contact and interacting with robots. A broader defi nition of the concept of the "social robot" seems useful for the future. It will be relevant wherever people without special qualifi cations are confronted with autonomous and intelligent machines.

From a social perspective, the increasing use of robots and AI systems is associated with both opportunities and risks.

Opportunities could lie, among other things, in the use of such systems as supporting tools, e.g. to take over mechanical repetitive activities (e.g. cleaning), to help with strenuous physical work (e.g. by exoskeletons), to enable people with disabilities or the elderly to lead an independent life (e.g. with driverless cars or service robots), or to counteract the detrimental effects of social isolation (e.g. by telepresence robots).

Risks lie, among other things, in a perceived loss of control, in uncertainties in dealing with autonomous machines, in new forms of psychological stress in highly automated work environments, in one-sided emotional ties to machines or in the threat to self-esteem when robots and AI systems take over competences from humans. As intelligent agents also simu- late social-communicative behaviour and users build some kind of emotional relationship with the machine, questions of susceptibility to emotional manipulation or a possible loss of interest in actual interpersonal relationships arise19.

In addition, risks are refl ected in reservations and fears that exist in the public towards autonomous technologies. In addition to fears of "being replaced" and, in particular, the substi- tution of socio-emotional ("human") competences, scepticism also prevails in society due to the fear of dependence on machines, de- It must be ensured that decisions cision-making by machines and the associa- made with the help of algorithmic ted dominance of algorithmic classifi cation processes and AI systems do not schemes (“fear of being categorized”). Today, discriminate against anyone. AI is increasingly used in so-called algorithmic prognosis and decision-making processes (e.g. predictive analytics). This means that large amounts of (historical) data are subjected to machine-aided statistical analysis and learning processes in order to identify possible correlations that help to categorise current events or predict future events. 38 Austrian Council on Robotics and Artifi cial Intelligence

These processes have enormous economic potential, as they can be used to forecast the future purchasing behaviour of customers and to adjust inventories accordingly, or to precisely predict and plan the maintenance and replacement of machine parts. This method provides important opportunities for medical applications, for example for diagnosing diseases such as certain types of cancer. The expectation in this fi eld is to have earlier and more accurate prognoses.

Whenever predictive analytics methods are used to predict individual behaviour or characteristics of people and these forecasts are used as a basis for decision-making, ethical questions might arise. In the United States, these methods are used to predict the probability of recidivism among criminal offenders20. A considerable number of companies already uses AI systems in HR management to analyse video-recorded job interviews and biogra- phical information of job candidates to draw conclusions on the personality and establish whether they are fi t for a certain position21. In addition, such predictive analysis methods are used to decide whether to grant insurance or consumer credits. There are several reasons why this is problematic: fi rst, the accuracy of this prognosis strongly depends on the quality of the training data. Second, the weighting of individual factors used in the prediction is often not (or no longer) comprehensible.

One of the risks is that existing stereotypes It is necessary to review the existing and prejudices against certain groups will be ethical and legal framework with regard reproduced in the data used by such technolo- to psychological, social and socio-cultural gies, leading to unfair and discriminatory deci- changes and, if necessary, to introduce sions22. Therefore, people's fears of being con- new regulations and standards which fronted with technologies which they do not guarantee the safety of use of robotics understand or their fears of becoming "trans- and AI. parent" due to tracking of personal data by robots and AI systems23 must be taken seriously.

The Council on Robotics and Artifi cial Intelligence urges stakeholders to prevent or mitigate the risks listed above by taking appropriate measures. It is also necessary to consider existing and new developments in robotics and AI with regard to psychological, social and socio-cultural effects for Austrians and to assess the associated opportunities and risks. To be able to do so, the Council on Robotics and Artifi cial Intelligence sees a need for further research.

In addition, there is also a clear need for legal regulation and the need to proactively incorporate appropriate proposals into European legislation. The Council therefore recommends that the existing ethical and legal framework be reviewed in the light of upcoming changes and, where appropriate, add new rules and standards to ensure the safe use of robotics and AI for individuals and society. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 39

The key problems are summarised below (cf. Borges, 2018):

It is generally assumed that the use of robotics and AI reduces the probability of errors and potential damage, e.g. in road traffi 24c . However, this must not lead to the neglect of ethical questions which rarely occur in practice, but are nevertheless of great signifi cance (such as the "dilemma situations" already mentioned above). This is a matter of maintaining the high ethical standards currently in place in Austria. Liability regulations must therefore be adapted to these new challenges. This does not mean that the principles of tort law are to be changed, but rather a modernisation of the rules and standards for the use of robotics and AI. Even if a robot acts autonomously, a person must assume liability for illegal behaviour. As of now, this person could be the holder, the owner, the user or the manufacturer. Under current law, the breach of law can only stem from the behaviour of these persons, but not from an error made by an autonomous system (liability cannot be transferred to autonomous systems).

It may therefore be necessary to adapt exis- It is essential to maintain the high ting regulations (legislation such as the "Ma- ethical standards in Austria by chinery Directive" or technical standards) with adapting the legal framework with regard to responsibilities and duties. Stricter regard to liability, responsibility and maintenance obligations could be introduced duties. or lawmakers may rely more strongly on product hazard and product liability regulations.

Decisions by autonomous systems need to be checked for compliance with legal and ethical standards. The Council recommends the development of appropriate certifi cation and auditing/compliance tools for robotics and AI technologies.

According to the General Data Protection Regulation (GDPR), no person may be subject to an automated decision which has legal or other signifi cant consequences for them, subject to certain exceptions (conclusion or performance of a contract, consent or legal regulation). However, the right to explanation and transparency does not include the obligation to dis- close the algorithm that was used.

Questions on appropriate tools and measures for ensuring the transparency of AI systems should be discussed at the European level.

Appropriate certiﬁ cation and Autonomous systems produce a vast amount auditing/compliance tools for of data that are of interest to manufacturers, robotics and AI technologies are to owners, drivers and third parties. Regarding be developed at European level. data usage rights, additional research and regulation is needed, for example, regarding personal data, where the principle of informa- tional self-determination pursuant to the GDPR applies. Data may only be processed with consent, in fulfi lment of a contract, on the basis of legal provisions or in case of a legitimate interests. Transparency must be ensured in any case (additional data processing is possible in compliance with "Privacy by Design and Default"). 40 Austrian Council on Robotics and Artifi cial Intelligence

Finally, the use of autonomous systems changes the standard of due diligence under criminal law. User’s duty of care will be alleviated, while manufacturers and holders have to assume a higher level of liability. From the Council's point of view, more precise regulation is needed. For example, the risk that in certain situations no one would be criminally responsible could be tackled with the concept of shared responsibility.

Although not a priority at present, the Council believes that the question of the legal personality of robots should be further pursued. However, for now it is more important to formulate rules and standards for the use of robots in order to determine liability issues.

5.4. Awareness Raising, Communication and Public Relations

Robotics and AI are topics that people have been concerning themselves with for a long time. Science fi ction has also contributed helped to achieve a high level of visibility. Due to the increasing performance of cutting-edge computer technology, which in connection with digital high-speed networks can process huge amounts of data in a very short time, a previously rather visionary picture of the future quickly becomes reality, either now or in the foreseeable future. However, very often only specialists understand the technical functio- nalities of AI. Many people do not have detailed knowledge of these technologies and are therefore not fully able to distinguish between reality and fi ction. This lack of understanding, combined with science fi ction information from the past and little objective information in the media, leads to doubts and fear among the general public. Thus it is not surprising that potential negative effects dominate discussion over chances and opportunities.

In order to be able to meaningfully shape the future with these technologies, it is urgently recommended to provide people with reliable and comprehensive information. This is the only way for citizens to assess the opportunities and risks of robotics and AI. Information should focus on meaningful future scenarios and on raising awareness for the social relevance of individuals’ deeds. We should strive for a way of life in which we as human beings can use their individual abilities and enjoy autonomy.

Robotics and AI create opportunities that Citizens should be able to form their allow people to spend more time applying opinion on the risks and opportunities human strengths and abilities. This includes of robotics and AI. Therefore, Austrian creative activities, solving unstructured tasks, citizens should be encouraged to interaction with others in need of empathy participate in the creation of the and social intelligence, and much more. country’s Robotics and AI Strategy to provide information and education, From the Council's point of view, it is indis- and thereby gain acceptance. pensable to create such images from that contrast science fi ction narratives in order generate public awareness and to enable meaningful dialogue. Shaping the Future of Austria with Robotics and Artifi cial Intelligence 41

The Council therefore recommends the broad participation of Austrian citizens in the development of the Austrian robotics and AI strategy to inform, educate and increase public acceptance. We recommend the "Montreal Declaration on Responsible AI”25 as reference.

Within the framework of citizen participation, measures must be developed as soon as possible which are suitable for bringing people into contact with robotics and AI technologies and informing them of their potential and risks. One of the fi rst measures could be, for example, to invite Austrian kindergartens and schools to visit research facilities26. Universities and research institutions must provide the resources for such visits. Creating a well-designed, up-to-date, interactive website at the Council on Robotics and Artifi cial Intelligence could be another measure for short-term implementation. Additional platforms are to be created to ensure that all stakeholders are heard and have the opportunity to participate from the beginning. 42 Austrian Council on Robotics and Artificial Intelligence Shaping the Future of Austria with Robotics and Artifi cial Intelligence 43

6. Strategy Process and Governance

The members of the Council on Robotics and Artifi cial Intelligence consider this white paper as a fi rst impulse for the development of an appropriate robotics and AI strategy for Austria. It is therefore not a fully fi nished concept and requires ongoing adaptation to incorporate new fi ndings and developments as well as to the results and outcomes of a broad public discourse with all relevant stakeholders. Due to the complexity of the problems and the dynamics in the fi elds of robotics and AI, long-term planning on the basis of complete information on the current and future situation is impossible from the Council's point of view.

Against the background of these infl uences, In view of the complexity and fastmoving it is not expedient to develop an all-encom- nature of the problem, it makes sense to passing strategy in the sense of a "big bang”. develop and implement an incremental Instead, the Council recommends incremen- strategy with institutionalised feedback tal strategy development and implementation loops, and continuous research and with institutionalised feedback loops, and technology monitoring. continuous research and technology monitoring (“strategising").

With the objectives mentioned above (1) to promote technological and industrial performance, (2) to prepare for socio-economic changes associated with robotics and AI, and (3) to ensure an appropriate ethical and legal framework, the route is clear. In the interest of rapid success and feedback, short-term measures to leverage potentials ("quick wins”) and urgently needed measures to prevent and mitigate negative consequences should be implemented as soon as possible.

The Council recommends a broad involve- In strategy development it is important ment of stakeholders from the all fi elds of ac- to ensure a large-scale involvement of Aus- tion (e.g. research, technology development trian stakeholders and intensive network- and production, business, education and ing with strategic partners and activities at qualifi cation, health, safety, housing, mobility) the European level. and forming strong networks with the strategic activities currently taking place at the European level (European AI Alliance and the AI High Level Expert Group).

The fundamental structural changes addressed in this document , for example in RTI, in connection with the digital transformation of the Austrian economy, the school and education system, questions of social security, etc., are to be initiated promptly by initiating appropriate institutional processes. 44 Austrian Council on Robotics and Artificial Intelligence

End Notes

1 e.g. in the Charter of Fundamental Rights of the 15 Cf. 14 See Pichler et al. (2017) European Union (2012) 16 Think, for example, of taxi drivers: with the help 2 Principle of shared responsibility, see for of a navigation system they no longer need to know example WEF (2015) where their destination is located to perform their job. 3 In values of the European Union in particular; the Communication of the European Commission (2018) 17 Cf. Urry (2017); Ho & MacDorman (2010); Mara & contains helpful references and sources and Appel (2015); Mori, MacDorman & Kageki (2012) announces a draft version of ethical guidelines for artificial intelligence by the end of 2018. 18 Cf. European Commission (2012)

4 IEEE Ethics initiative ( https://standards.ieee.org/ 19 Cf. Scheutz (2011) develop/indconn/ec/autonomous_systems.html ) 20 Cf. Perset et al. (2018) 5 Cf. 5 Cf. by Schomberg (2011) 21 Cf. 21 See The Economist (2018); Accenture, for 6 Cf. Owen, Bessant & Heintz (2013) example, uses HireVue, an AI program that analyzes video interviews with to assess whether they match 7 The literature provides numerous references to the position; Johnson & Johnson uses the HiredScore ethics in research, see for example the University of AI system to evaluate candidates. Vienna. (o.D.) or by Unger & Narimani (2014) 22 Cf. Bryson et al. (2017) and Perset et al. (2018) 8 The term "robot" with its Czech origin in the word "robota" first appeared inalmost 100 years ago in Karel 23 Cf. Bryson et al. (2017) Čapek’s play R.U.R., in which artificially created, human-like machines revolt against their enslavement. 24 Cf. 24 Eisenberger, Lachmayer & Eisenberger (2017) 9 “[Autonomous] Cognitive Systems are systems that perceive, understand, learn and develop through 25 Montreal Declaration on Responsible AI ( https:// individual or social interaction with their environment. www.declarationmontreal-iaresponsable.com ) [...] As a scientific discipline, Cognitive Systems seeks to provide an enabling technology for robotics and 26 For example, the Institute for Automation and automation, natural language understanding, man-ma- Control Technology at TU Wien and the Department chine interaction and complex real-world systems of Industrial Engineering at the Vienna University of (Europäische Union, 2002) und “[The objective is] to Applied Sciences already offer´ robotics workshops construct physically instantiated or embodied systems for schools where children can experience robots that can perceive, understand (the semantics of infor- and also program them. Similar efforts are being mation conveyed through their perceptual input) and made by the PRIA Institute ( https://pria.at/ ) and interact with their environment, and evolve in order to other institutions in Austria. achieve human-like performance in activities requiring context-(situation and task) specific knowledge.” (US Department of Defense, 2015) Note: REGULATION (EU) 2016/679 OF THE EUROPEAN 10 AI systems should store all information on PARLIAMENT AND OF THE COUNCIL of 27 algorithms and decision-making rules (on ethical April 2016 on the protection of natural persons questions) in an "ethics black box" in order to ensure with regard to the processing of personal data and that decisions can be "explained". See e.g. UNI Global on the free movement of such data, and repealing Union (2017) Directive 95/46/EC (General Data Protection Regu- lation) 11 Cf. Wachter, Mittelstadt & Russell, 2017

12 Cf. European Parliament (2017)

13 Cf. European Parliament (2016)

14 Cf. 14 See Pichler et al. (2017); Čas, Rose & Schüttler (2017) Shaping the Future of Austria with Robotics and Artificial Intelligence 45

Glossary

Actuators Actuators convert electrical signals into mechanical motion or other physical quantities.

Algorithm Specification on how to solve a task.

Autonomous driving Using a driverless transport system with autonomous behaviour.

Drone Unmanned aircraft operated autonomously or controlled from the ground.

Exoskeleton External support structure for an organism that can also be used to increase performance with the support of technology.

Humanoid robot Robots with human-like shape or characteristics.

Collaborative Robot Industrial robot that works together with people, i.e. in the same workspace.

Sandbox An isolated area where actions have no effect on the external environment.

Science Fiction A narrative that depicts scientific and technical speculations on a distant future.

Software agents A computer program that is capable of certain (specialised) independent and self-dynamic (autonomous) behaviour.

Strategising Includes all practical measures taken by individuals to develop long-term goals, plans and actions.

Testbed Scientific platform for experiments and research. 46 Austrian Council on Robotics and Artificial Intelligence

List of References

Autor, D. H. (2015). Why Are There Still So Many Jobs? Nedelkoska, L. & Quintini G. (2018). Automation, skills The History and Future of Workplace Automation. use and training. OECD Social, Employment and Migra- Journal of Economic Perspectives, 29, 3–30. tion Working Papers. No. 202. OECD Publishing. Paris. https://doi.org/10.1257/jep.29.3.3 https://doi.org/10.1787/2e2f4eea-en

Borges, G. (2018). Legal framework for autonomous Owen, R., Bessant, J., & Heintz, M. (2013). Responsible Systems. NJW 20.. 977 ff. Innovation. Chichester, UK: John Wiley & Sons, Ltd.

Brynjolfsson, E., & McAfee, A. (2014). Race against the Perset, K., Nishigata, N. & Carblanc, A. (2018). Artificial machine: How the digital revolution is accelerating Intelligence in Society (Phase1). OECD Document innovation, driving productivity, and irreversibly JT03430927. DSTI/CDEP(2018)9. Directorate for transforming employment and the economy. Lexington, Science, Technology and Innovation. Massachusetts: Digital Frontier Press. Pichler, A., Wögerer, C., Fritz, G., Vincze, M., Steinbauer, Bryson, J. J., Diamantis, M. E., & Grant, T. D. (2017). Of, G., Hofbaur, M., Kubinger, W., Gspandl, S.,Müller, A. for, and by the people: the legal lacuna of synthetic & Piater, J. (2017). Potential Assessment “Robotics in persons. Artificial Intelligence and Law, 25, 273–291. Austria”. Commissioned by: BMVIT: Association for https://doi.org/10.1007/s10506-017-9214-9 Metrology, Automation and Robotics technology (GMAR) / Expert Group for Robotics. Čas, J., Rose, G., & Shaker, L. (2017). Robotics in Austria. Short study - Development perspectives and Scheutz, M. (2011). The Inherent Dangers of Unidirec- political challenges. On behalf of BMVIT: Institute for tional Emotional Bonds between Humans and Social Technology Assessment of the Austrian Academy of Robots. Robot ethics: the ethical and social implicati- Sciences. Project Report No. 2017-03 | ISSN: 1819- ons of robotics. Ed. Lin, Patrick. Cambridge, 1320 ISSN->: 1818-6556. Mass.: MIT Press

Eisenberger, I., Lachmayer, K., & Eisenberger, G. (Eds.). By Unger, H., & Narimani, P. (Eds.). (2014). Ethics of (2017). Autonomous driving and law. Vienna: research in qualitative research: reflexivity, perspecti- Manz'sche Verlags- und Universitätsbuchhandlung. ves, positions. Springer-Verlag.

Ford, M. (2015). The rise of the robots: Technology and Wachter, S., Mittelstadt, B., & Russell, C. (2017). the threat of mass unemployment. A One-world book. Counterfactual Explanations Without Opening the London: Oneworld. Black Box: Automated Decisions and the GDPR. SSRN Electronic Journal. Advance online publication. Ms., C.. B.., & Osborne, C. A.. (2017). The future of https://doi.org/10.2139/ssrn.3063289 employment: How susceptible are jobs to computerisation? Technological Forecasting and Soci- al Change, 114, 254–280. https://doi.org/10.1016/j.techfore.2016.08.019

Ho, C.-C., & MacDorman, K. F. (2010). Revisiting the uncanny valley theory: Developing and validating an alternative to the Godspeed indices. Computers in Human Behavior, 26, 1508–1518. https://doi. org/10.1016/j.chb.2010.05.015

Mara, M., & Appel, M. (2015). Robots in the creepy ditch: Why human-like machines are often uncanny to us. In-Mind. Downloaded from: http://de.in-mind.org/article/roboter-im-grusel- graben-warum-uns-menschenaehnliche-maschi- nen-oft-unheimlich-sind

Mori, M., MacDorman, K., & Kageki, N. (2012). The Uncanny Valley [From the Field]. IEEE Robotics & Automation Magazine, 19, 98–100. https://doi.org/10.1109/MRA.2012.2192811 Shaping the Future of Austria with Robotics and Artificial Intelligence 47

Online Resources

Last accessed on 01.10.2018 The Economist. (2018). Non-tech businesses are beginning to use artificial intelligence at scale (GrAIt Brundage, M. et al. (2018). The Malicious Use of expectations). Downloaded from https://www.econo- Artificial Intelligence: Forecasting, Prevention, and mist.uni-global-union_-artificial-intelligence.pdf Mitigation. Downloaded from https: //arxiv.org/ftp/ arxiv/papers/1802/1802.07228.pdf UNI Global Union. (2017). The 10 most important principles for ethical artificial intelligence. Source European Commission (2012). Special Eurobaro- me- link: http://www.thefutureworldofwork.org/me- ter 382: Public attitudes towards robots. Downloaded dia/35484/ uni-global-union_-artificial-intelligence. from http://ec.europa.eu/public_opinion/archives/ pdf ebs/ebs_382_en.pdf University of Vienna. (o.D.). Research ethical principles. European Commission (2018). Artificial intelligence for Source link: https://pflegewissenschaft.univie.ac.at/ Europe. Downloaded from https://ec.europa.eu/ fileadmin/user_upload/inst_pflegewiss/Homepage_ transparency/regdoc rep/1/2018/EN/COM-2018- neu/Home/Ethik/Forschungsethische_Prinzipien.pdf 237-F1-EN-MAIN-PART-1.PDF Urry, H. L. (20., 16.. August). The Uncanny Valley and European Union. (2002). Cognitive systems. Retrieved its Mechanisms [blog entry]. Source link: http:// from https://cordis.europa.eu/programme/ sites.tufts.edu/emotiononthebrain/2017/08/16/ rcn/8152_en.html theuncanny-valley-and-its-mechanisms/

European Union. (2012). Charter of Fundamental US Department of Defense. (2015). Unmanned Sys- Rights of the European Union. Downloaded from tems Integrated Roadmap FY2011-2036. Source link: https:// eur-lex.europa.eu/legal-content/DE/TXT/? https://fas.org/irp/program/collect/usroadmap2011. uri=CE- LEX:12012P/TXT pdf

European Parliament. (2016). European Civil Law Ru- From Schomberg, R. (2011). Definition of Responsible les in Robotics. Downloaded from http://www.europarl. innovation. Downloaded from https://renevonschom- europa.eu/RegData/etudes/STUD/2016/571379/ berg.wordpress.com/definition-of-responsi- bleinno- IPOL_STU(2016)571379_EN.pdf vation/

European Parliament. (2017). Civil law regulations in WEF. (2015). Shared Responsibility: A New Paradigm the field of robotics. Downloaded from http://www. for Supply Chains. Source link http://www3.weforum. europarl.europa.eu/sides/getDoc.do?pubRef=-// org/docs/WEF_GAC_Supply_ Chains_%20A_ EP//TEXT+TA+P8-TA-2017-0051+0+DOC+XML+ New_Paradigm_2015.pdf V0//DE WEF. (2018). The Future of Jobs Report 2018 Source IHS. (2017). Digitisation endangers far fewer jobs link: http://www3.weforum.org/docs/WEF_Future_ than feared [press release]. Retrieved from of_Jobs_2018.pdf https://www.ihs.ac.at/fileadmin/public/BR_Files/ user_upload/ 20170412_PK_Digitalization.pdf

Mittelstadt, B. D., Allo, P., Taddeo, M., Wachter, S., & Floridi, L. (2016). The ethics of algorithms: Mapping the debate. Big Data & Society, 3(2), 205395171667967. https://doi.org/10.1177/2053951716679679

Montreal Declaration on Responsible AI (o.D.). Calling from https://www.declarationmontrealiaresponsable. com

Rohde, N. (2018). Quality criteria for algorithmic processes. Downloaded from https://www.bertels- mann-stiftung.de/en/publikationen/publikation/did/ guetekriterien-fuer-algorithmic-processes/ 48 Austrian Council on Robotics and Artificial Intelligence

Imprint (German)

Editor and Media Owner Austrian Council on Robotics and Artificial Intelligence 1010 Vienna, Pestalozzigasse 4 www.acrai.at

Design Haemmerle & Luger www.haemmerle-luger.com

Illustration Alexandar Savić